1 //===- CodeExtractor.cpp - Pull code region into a new function -----------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This file implements the interface to tear out a code region, such as an 10 // individual loop or a parallel section, into a new function, replacing it with 11 // a call to the new function. 12 // 13 //===----------------------------------------------------------------------===// 14 15 #include "llvm/Transforms/Utils/CodeExtractor.h" 16 #include "llvm/ADT/ArrayRef.h" 17 #include "llvm/ADT/DenseMap.h" 18 #include "llvm/ADT/Optional.h" 19 #include "llvm/ADT/STLExtras.h" 20 #include "llvm/ADT/SetVector.h" 21 #include "llvm/ADT/SmallPtrSet.h" 22 #include "llvm/ADT/SmallVector.h" 23 #include "llvm/Analysis/AssumptionCache.h" 24 #include "llvm/Analysis/BlockFrequencyInfo.h" 25 #include "llvm/Analysis/BlockFrequencyInfoImpl.h" 26 #include "llvm/Analysis/BranchProbabilityInfo.h" 27 #include "llvm/Analysis/LoopInfo.h" 28 #include "llvm/IR/Argument.h" 29 #include "llvm/IR/Attributes.h" 30 #include "llvm/IR/BasicBlock.h" 31 #include "llvm/IR/CFG.h" 32 #include "llvm/IR/Constant.h" 33 #include "llvm/IR/Constants.h" 34 #include "llvm/IR/DIBuilder.h" 35 #include "llvm/IR/DataLayout.h" 36 #include "llvm/IR/DebugInfoMetadata.h" 37 #include "llvm/IR/DerivedTypes.h" 38 #include "llvm/IR/Dominators.h" 39 #include "llvm/IR/Function.h" 40 #include "llvm/IR/GlobalValue.h" 41 #include "llvm/IR/InstIterator.h" 42 #include "llvm/IR/InstrTypes.h" 43 #include "llvm/IR/Instruction.h" 44 #include "llvm/IR/Instructions.h" 45 #include "llvm/IR/IntrinsicInst.h" 46 #include "llvm/IR/Intrinsics.h" 47 #include "llvm/IR/LLVMContext.h" 48 #include "llvm/IR/MDBuilder.h" 49 #include "llvm/IR/Module.h" 50 #include "llvm/IR/PatternMatch.h" 51 #include "llvm/IR/Type.h" 52 #include "llvm/IR/User.h" 53 #include "llvm/IR/Value.h" 54 #include "llvm/IR/Verifier.h" 55 #include "llvm/Pass.h" 56 #include "llvm/Support/BlockFrequency.h" 57 #include "llvm/Support/BranchProbability.h" 58 #include "llvm/Support/Casting.h" 59 #include "llvm/Support/CommandLine.h" 60 #include "llvm/Support/Debug.h" 61 #include "llvm/Support/ErrorHandling.h" 62 #include "llvm/Support/raw_ostream.h" 63 #include "llvm/Transforms/Utils/BasicBlockUtils.h" 64 #include "llvm/Transforms/Utils/Local.h" 65 #include <cassert> 66 #include <cstdint> 67 #include <iterator> 68 #include <map> 69 #include <set> 70 #include <utility> 71 #include <vector> 72 73 using namespace llvm; 74 using namespace llvm::PatternMatch; 75 using ProfileCount = Function::ProfileCount; 76 77 #define DEBUG_TYPE "code-extractor" 78 79 // Provide a command-line option to aggregate function arguments into a struct 80 // for functions produced by the code extractor. This is useful when converting 81 // extracted functions to pthread-based code, as only one argument (void*) can 82 // be passed in to pthread_create(). 83 static cl::opt<bool> 84 AggregateArgsOpt("aggregate-extracted-args", cl::Hidden, 85 cl::desc("Aggregate arguments to code-extracted functions")); 86 87 /// Test whether a block is valid for extraction. 88 static bool isBlockValidForExtraction(const BasicBlock &BB, 89 const SetVector<BasicBlock *> &Result, 90 bool AllowVarArgs, bool AllowAlloca) { 91 // taking the address of a basic block moved to another function is illegal 92 if (BB.hasAddressTaken()) 93 return false; 94 95 // don't hoist code that uses another basicblock address, as it's likely to 96 // lead to unexpected behavior, like cross-function jumps 97 SmallPtrSet<User const *, 16> Visited; 98 SmallVector<User const *, 16> ToVisit; 99 100 for (Instruction const &Inst : BB) 101 ToVisit.push_back(&Inst); 102 103 while (!ToVisit.empty()) { 104 User const *Curr = ToVisit.pop_back_val(); 105 if (!Visited.insert(Curr).second) 106 continue; 107 if (isa<BlockAddress const>(Curr)) 108 return false; // even a reference to self is likely to be not compatible 109 110 if (isa<Instruction>(Curr) && cast<Instruction>(Curr)->getParent() != &BB) 111 continue; 112 113 for (auto const &U : Curr->operands()) { 114 if (auto *UU = dyn_cast<User>(U)) 115 ToVisit.push_back(UU); 116 } 117 } 118 119 // If explicitly requested, allow vastart and alloca. For invoke instructions 120 // verify that extraction is valid. 121 for (BasicBlock::const_iterator I = BB.begin(), E = BB.end(); I != E; ++I) { 122 if (isa<AllocaInst>(I)) { 123 if (!AllowAlloca) 124 return false; 125 continue; 126 } 127 128 if (const auto *II = dyn_cast<InvokeInst>(I)) { 129 // Unwind destination (either a landingpad, catchswitch, or cleanuppad) 130 // must be a part of the subgraph which is being extracted. 131 if (auto *UBB = II->getUnwindDest()) 132 if (!Result.count(UBB)) 133 return false; 134 continue; 135 } 136 137 // All catch handlers of a catchswitch instruction as well as the unwind 138 // destination must be in the subgraph. 139 if (const auto *CSI = dyn_cast<CatchSwitchInst>(I)) { 140 if (auto *UBB = CSI->getUnwindDest()) 141 if (!Result.count(UBB)) 142 return false; 143 for (auto *HBB : CSI->handlers()) 144 if (!Result.count(const_cast<BasicBlock*>(HBB))) 145 return false; 146 continue; 147 } 148 149 // Make sure that entire catch handler is within subgraph. It is sufficient 150 // to check that catch return's block is in the list. 151 if (const auto *CPI = dyn_cast<CatchPadInst>(I)) { 152 for (const auto *U : CPI->users()) 153 if (const auto *CRI = dyn_cast<CatchReturnInst>(U)) 154 if (!Result.count(const_cast<BasicBlock*>(CRI->getParent()))) 155 return false; 156 continue; 157 } 158 159 // And do similar checks for cleanup handler - the entire handler must be 160 // in subgraph which is going to be extracted. For cleanup return should 161 // additionally check that the unwind destination is also in the subgraph. 162 if (const auto *CPI = dyn_cast<CleanupPadInst>(I)) { 163 for (const auto *U : CPI->users()) 164 if (const auto *CRI = dyn_cast<CleanupReturnInst>(U)) 165 if (!Result.count(const_cast<BasicBlock*>(CRI->getParent()))) 166 return false; 167 continue; 168 } 169 if (const auto *CRI = dyn_cast<CleanupReturnInst>(I)) { 170 if (auto *UBB = CRI->getUnwindDest()) 171 if (!Result.count(UBB)) 172 return false; 173 continue; 174 } 175 176 if (const CallInst *CI = dyn_cast<CallInst>(I)) { 177 if (const Function *F = CI->getCalledFunction()) { 178 auto IID = F->getIntrinsicID(); 179 if (IID == Intrinsic::vastart) { 180 if (AllowVarArgs) 181 continue; 182 else 183 return false; 184 } 185 186 // Currently, we miscompile outlined copies of eh_typid_for. There are 187 // proposals for fixing this in llvm.org/PR39545. 188 if (IID == Intrinsic::eh_typeid_for) 189 return false; 190 } 191 } 192 } 193 194 return true; 195 } 196 197 /// Build a set of blocks to extract if the input blocks are viable. 198 static SetVector<BasicBlock *> 199 buildExtractionBlockSet(ArrayRef<BasicBlock *> BBs, DominatorTree *DT, 200 bool AllowVarArgs, bool AllowAlloca) { 201 assert(!BBs.empty() && "The set of blocks to extract must be non-empty"); 202 SetVector<BasicBlock *> Result; 203 204 // Loop over the blocks, adding them to our set-vector, and aborting with an 205 // empty set if we encounter invalid blocks. 206 for (BasicBlock *BB : BBs) { 207 // If this block is dead, don't process it. 208 if (DT && !DT->isReachableFromEntry(BB)) 209 continue; 210 211 if (!Result.insert(BB)) 212 llvm_unreachable("Repeated basic blocks in extraction input"); 213 } 214 215 LLVM_DEBUG(dbgs() << "Region front block: " << Result.front()->getName() 216 << '\n'); 217 218 for (auto *BB : Result) { 219 if (!isBlockValidForExtraction(*BB, Result, AllowVarArgs, AllowAlloca)) 220 return {}; 221 222 // Make sure that the first block is not a landing pad. 223 if (BB == Result.front()) { 224 if (BB->isEHPad()) { 225 LLVM_DEBUG(dbgs() << "The first block cannot be an unwind block\n"); 226 return {}; 227 } 228 continue; 229 } 230 231 // All blocks other than the first must not have predecessors outside of 232 // the subgraph which is being extracted. 233 for (auto *PBB : predecessors(BB)) 234 if (!Result.count(PBB)) { 235 LLVM_DEBUG(dbgs() << "No blocks in this region may have entries from " 236 "outside the region except for the first block!\n" 237 << "Problematic source BB: " << BB->getName() << "\n" 238 << "Problematic destination BB: " << PBB->getName() 239 << "\n"); 240 return {}; 241 } 242 } 243 244 return Result; 245 } 246 247 CodeExtractor::CodeExtractor(ArrayRef<BasicBlock *> BBs, DominatorTree *DT, 248 bool AggregateArgs, BlockFrequencyInfo *BFI, 249 BranchProbabilityInfo *BPI, AssumptionCache *AC, 250 bool AllowVarArgs, bool AllowAlloca, 251 std::string Suffix) 252 : DT(DT), AggregateArgs(AggregateArgs || AggregateArgsOpt), BFI(BFI), 253 BPI(BPI), AC(AC), AllowVarArgs(AllowVarArgs), 254 Blocks(buildExtractionBlockSet(BBs, DT, AllowVarArgs, AllowAlloca)), 255 Suffix(Suffix) {} 256 257 CodeExtractor::CodeExtractor(DominatorTree &DT, Loop &L, bool AggregateArgs, 258 BlockFrequencyInfo *BFI, 259 BranchProbabilityInfo *BPI, AssumptionCache *AC, 260 std::string Suffix) 261 : DT(&DT), AggregateArgs(AggregateArgs || AggregateArgsOpt), BFI(BFI), 262 BPI(BPI), AC(AC), AllowVarArgs(false), 263 Blocks(buildExtractionBlockSet(L.getBlocks(), &DT, 264 /* AllowVarArgs */ false, 265 /* AllowAlloca */ false)), 266 Suffix(Suffix) {} 267 268 /// definedInRegion - Return true if the specified value is defined in the 269 /// extracted region. 270 static bool definedInRegion(const SetVector<BasicBlock *> &Blocks, Value *V) { 271 if (Instruction *I = dyn_cast<Instruction>(V)) 272 if (Blocks.count(I->getParent())) 273 return true; 274 return false; 275 } 276 277 /// definedInCaller - Return true if the specified value is defined in the 278 /// function being code extracted, but not in the region being extracted. 279 /// These values must be passed in as live-ins to the function. 280 static bool definedInCaller(const SetVector<BasicBlock *> &Blocks, Value *V) { 281 if (isa<Argument>(V)) return true; 282 if (Instruction *I = dyn_cast<Instruction>(V)) 283 if (!Blocks.count(I->getParent())) 284 return true; 285 return false; 286 } 287 288 static BasicBlock *getCommonExitBlock(const SetVector<BasicBlock *> &Blocks) { 289 BasicBlock *CommonExitBlock = nullptr; 290 auto hasNonCommonExitSucc = [&](BasicBlock *Block) { 291 for (auto *Succ : successors(Block)) { 292 // Internal edges, ok. 293 if (Blocks.count(Succ)) 294 continue; 295 if (!CommonExitBlock) { 296 CommonExitBlock = Succ; 297 continue; 298 } 299 if (CommonExitBlock != Succ) 300 return true; 301 } 302 return false; 303 }; 304 305 if (any_of(Blocks, hasNonCommonExitSucc)) 306 return nullptr; 307 308 return CommonExitBlock; 309 } 310 311 CodeExtractorAnalysisCache::CodeExtractorAnalysisCache(Function &F) { 312 for (BasicBlock &BB : F) { 313 for (Instruction &II : BB.instructionsWithoutDebug()) 314 if (auto *AI = dyn_cast<AllocaInst>(&II)) 315 Allocas.push_back(AI); 316 317 findSideEffectInfoForBlock(BB); 318 } 319 } 320 321 void CodeExtractorAnalysisCache::findSideEffectInfoForBlock(BasicBlock &BB) { 322 for (Instruction &II : BB.instructionsWithoutDebug()) { 323 unsigned Opcode = II.getOpcode(); 324 Value *MemAddr = nullptr; 325 switch (Opcode) { 326 case Instruction::Store: 327 case Instruction::Load: { 328 if (Opcode == Instruction::Store) { 329 StoreInst *SI = cast<StoreInst>(&II); 330 MemAddr = SI->getPointerOperand(); 331 } else { 332 LoadInst *LI = cast<LoadInst>(&II); 333 MemAddr = LI->getPointerOperand(); 334 } 335 // Global variable can not be aliased with locals. 336 if (dyn_cast<Constant>(MemAddr)) 337 break; 338 Value *Base = MemAddr->stripInBoundsConstantOffsets(); 339 if (!isa<AllocaInst>(Base)) { 340 SideEffectingBlocks.insert(&BB); 341 return; 342 } 343 BaseMemAddrs[&BB].insert(Base); 344 break; 345 } 346 default: { 347 IntrinsicInst *IntrInst = dyn_cast<IntrinsicInst>(&II); 348 if (IntrInst) { 349 if (IntrInst->isLifetimeStartOrEnd()) 350 break; 351 SideEffectingBlocks.insert(&BB); 352 return; 353 } 354 // Treat all the other cases conservatively if it has side effects. 355 if (II.mayHaveSideEffects()) { 356 SideEffectingBlocks.insert(&BB); 357 return; 358 } 359 } 360 } 361 } 362 } 363 364 bool CodeExtractorAnalysisCache::doesBlockContainClobberOfAddr( 365 BasicBlock &BB, AllocaInst *Addr) const { 366 if (SideEffectingBlocks.count(&BB)) 367 return true; 368 auto It = BaseMemAddrs.find(&BB); 369 if (It != BaseMemAddrs.end()) 370 return It->second.count(Addr); 371 return false; 372 } 373 374 bool CodeExtractor::isLegalToShrinkwrapLifetimeMarkers( 375 const CodeExtractorAnalysisCache &CEAC, Instruction *Addr) const { 376 AllocaInst *AI = cast<AllocaInst>(Addr->stripInBoundsConstantOffsets()); 377 Function *Func = (*Blocks.begin())->getParent(); 378 for (BasicBlock &BB : *Func) { 379 if (Blocks.count(&BB)) 380 continue; 381 if (CEAC.doesBlockContainClobberOfAddr(BB, AI)) 382 return false; 383 } 384 return true; 385 } 386 387 BasicBlock * 388 CodeExtractor::findOrCreateBlockForHoisting(BasicBlock *CommonExitBlock) { 389 BasicBlock *SinglePredFromOutlineRegion = nullptr; 390 assert(!Blocks.count(CommonExitBlock) && 391 "Expect a block outside the region!"); 392 for (auto *Pred : predecessors(CommonExitBlock)) { 393 if (!Blocks.count(Pred)) 394 continue; 395 if (!SinglePredFromOutlineRegion) { 396 SinglePredFromOutlineRegion = Pred; 397 } else if (SinglePredFromOutlineRegion != Pred) { 398 SinglePredFromOutlineRegion = nullptr; 399 break; 400 } 401 } 402 403 if (SinglePredFromOutlineRegion) 404 return SinglePredFromOutlineRegion; 405 406 #ifndef NDEBUG 407 auto getFirstPHI = [](BasicBlock *BB) { 408 BasicBlock::iterator I = BB->begin(); 409 PHINode *FirstPhi = nullptr; 410 while (I != BB->end()) { 411 PHINode *Phi = dyn_cast<PHINode>(I); 412 if (!Phi) 413 break; 414 if (!FirstPhi) { 415 FirstPhi = Phi; 416 break; 417 } 418 } 419 return FirstPhi; 420 }; 421 // If there are any phi nodes, the single pred either exists or has already 422 // be created before code extraction. 423 assert(!getFirstPHI(CommonExitBlock) && "Phi not expected"); 424 #endif 425 426 BasicBlock *NewExitBlock = CommonExitBlock->splitBasicBlock( 427 CommonExitBlock->getFirstNonPHI()->getIterator()); 428 429 for (auto PI = pred_begin(CommonExitBlock), PE = pred_end(CommonExitBlock); 430 PI != PE;) { 431 BasicBlock *Pred = *PI++; 432 if (Blocks.count(Pred)) 433 continue; 434 Pred->getTerminator()->replaceUsesOfWith(CommonExitBlock, NewExitBlock); 435 } 436 // Now add the old exit block to the outline region. 437 Blocks.insert(CommonExitBlock); 438 return CommonExitBlock; 439 } 440 441 // Find the pair of life time markers for address 'Addr' that are either 442 // defined inside the outline region or can legally be shrinkwrapped into the 443 // outline region. If there are not other untracked uses of the address, return 444 // the pair of markers if found; otherwise return a pair of nullptr. 445 CodeExtractor::LifetimeMarkerInfo 446 CodeExtractor::getLifetimeMarkers(const CodeExtractorAnalysisCache &CEAC, 447 Instruction *Addr, 448 BasicBlock *ExitBlock) const { 449 LifetimeMarkerInfo Info; 450 451 for (User *U : Addr->users()) { 452 IntrinsicInst *IntrInst = dyn_cast<IntrinsicInst>(U); 453 if (IntrInst) { 454 // We don't model addresses with multiple start/end markers, but the 455 // markers do not need to be in the region. 456 if (IntrInst->getIntrinsicID() == Intrinsic::lifetime_start) { 457 if (Info.LifeStart) 458 return {}; 459 Info.LifeStart = IntrInst; 460 continue; 461 } 462 if (IntrInst->getIntrinsicID() == Intrinsic::lifetime_end) { 463 if (Info.LifeEnd) 464 return {}; 465 Info.LifeEnd = IntrInst; 466 continue; 467 } 468 // At this point, permit debug uses outside of the region. 469 // This is fixed in a later call to fixupDebugInfoPostExtraction(). 470 if (isa<DbgInfoIntrinsic>(IntrInst)) 471 continue; 472 } 473 // Find untracked uses of the address, bail. 474 if (!definedInRegion(Blocks, U)) 475 return {}; 476 } 477 478 if (!Info.LifeStart || !Info.LifeEnd) 479 return {}; 480 481 Info.SinkLifeStart = !definedInRegion(Blocks, Info.LifeStart); 482 Info.HoistLifeEnd = !definedInRegion(Blocks, Info.LifeEnd); 483 // Do legality check. 484 if ((Info.SinkLifeStart || Info.HoistLifeEnd) && 485 !isLegalToShrinkwrapLifetimeMarkers(CEAC, Addr)) 486 return {}; 487 488 // Check to see if we have a place to do hoisting, if not, bail. 489 if (Info.HoistLifeEnd && !ExitBlock) 490 return {}; 491 492 return Info; 493 } 494 495 void CodeExtractor::findAllocas(const CodeExtractorAnalysisCache &CEAC, 496 ValueSet &SinkCands, ValueSet &HoistCands, 497 BasicBlock *&ExitBlock) const { 498 Function *Func = (*Blocks.begin())->getParent(); 499 ExitBlock = getCommonExitBlock(Blocks); 500 501 auto moveOrIgnoreLifetimeMarkers = 502 [&](const LifetimeMarkerInfo &LMI) -> bool { 503 if (!LMI.LifeStart) 504 return false; 505 if (LMI.SinkLifeStart) { 506 LLVM_DEBUG(dbgs() << "Sinking lifetime.start: " << *LMI.LifeStart 507 << "\n"); 508 SinkCands.insert(LMI.LifeStart); 509 } 510 if (LMI.HoistLifeEnd) { 511 LLVM_DEBUG(dbgs() << "Hoisting lifetime.end: " << *LMI.LifeEnd << "\n"); 512 HoistCands.insert(LMI.LifeEnd); 513 } 514 return true; 515 }; 516 517 // Look up allocas in the original function in CodeExtractorAnalysisCache, as 518 // this is much faster than walking all the instructions. 519 for (AllocaInst *AI : CEAC.getAllocas()) { 520 BasicBlock *BB = AI->getParent(); 521 if (Blocks.count(BB)) 522 continue; 523 524 // As a prior call to extractCodeRegion() may have shrinkwrapped the alloca, 525 // check whether it is actually still in the original function. 526 Function *AIFunc = BB->getParent(); 527 if (AIFunc != Func) 528 continue; 529 530 LifetimeMarkerInfo MarkerInfo = getLifetimeMarkers(CEAC, AI, ExitBlock); 531 bool Moved = moveOrIgnoreLifetimeMarkers(MarkerInfo); 532 if (Moved) { 533 LLVM_DEBUG(dbgs() << "Sinking alloca: " << *AI << "\n"); 534 SinkCands.insert(AI); 535 continue; 536 } 537 538 // Find bitcasts in the outlined region that have lifetime marker users 539 // outside that region. Replace the lifetime marker use with an 540 // outside region bitcast to avoid unnecessary alloca/reload instructions 541 // and extra lifetime markers. 542 SmallVector<Instruction *, 2> LifetimeBitcastUsers; 543 for (User *U : AI->users()) { 544 if (!definedInRegion(Blocks, U)) 545 continue; 546 547 if (U->stripInBoundsConstantOffsets() != AI) 548 continue; 549 550 Instruction *Bitcast = cast<Instruction>(U); 551 for (User *BU : Bitcast->users()) { 552 IntrinsicInst *IntrInst = dyn_cast<IntrinsicInst>(BU); 553 if (!IntrInst) 554 continue; 555 556 if (!IntrInst->isLifetimeStartOrEnd()) 557 continue; 558 559 if (definedInRegion(Blocks, IntrInst)) 560 continue; 561 562 LLVM_DEBUG(dbgs() << "Replace use of extracted region bitcast" 563 << *Bitcast << " in out-of-region lifetime marker " 564 << *IntrInst << "\n"); 565 LifetimeBitcastUsers.push_back(IntrInst); 566 } 567 } 568 569 for (Instruction *I : LifetimeBitcastUsers) { 570 Module *M = AIFunc->getParent(); 571 LLVMContext &Ctx = M->getContext(); 572 auto *Int8PtrTy = Type::getInt8PtrTy(Ctx); 573 CastInst *CastI = 574 CastInst::CreatePointerCast(AI, Int8PtrTy, "lt.cast", I); 575 I->replaceUsesOfWith(I->getOperand(1), CastI); 576 } 577 578 // Follow any bitcasts. 579 SmallVector<Instruction *, 2> Bitcasts; 580 SmallVector<LifetimeMarkerInfo, 2> BitcastLifetimeInfo; 581 for (User *U : AI->users()) { 582 if (U->stripInBoundsConstantOffsets() == AI) { 583 Instruction *Bitcast = cast<Instruction>(U); 584 LifetimeMarkerInfo LMI = getLifetimeMarkers(CEAC, Bitcast, ExitBlock); 585 if (LMI.LifeStart) { 586 Bitcasts.push_back(Bitcast); 587 BitcastLifetimeInfo.push_back(LMI); 588 continue; 589 } 590 } 591 592 // Found unknown use of AI. 593 if (!definedInRegion(Blocks, U)) { 594 Bitcasts.clear(); 595 break; 596 } 597 } 598 599 // Either no bitcasts reference the alloca or there are unknown uses. 600 if (Bitcasts.empty()) 601 continue; 602 603 LLVM_DEBUG(dbgs() << "Sinking alloca (via bitcast): " << *AI << "\n"); 604 SinkCands.insert(AI); 605 for (unsigned I = 0, E = Bitcasts.size(); I != E; ++I) { 606 Instruction *BitcastAddr = Bitcasts[I]; 607 const LifetimeMarkerInfo &LMI = BitcastLifetimeInfo[I]; 608 assert(LMI.LifeStart && 609 "Unsafe to sink bitcast without lifetime markers"); 610 moveOrIgnoreLifetimeMarkers(LMI); 611 if (!definedInRegion(Blocks, BitcastAddr)) { 612 LLVM_DEBUG(dbgs() << "Sinking bitcast-of-alloca: " << *BitcastAddr 613 << "\n"); 614 SinkCands.insert(BitcastAddr); 615 } 616 } 617 } 618 } 619 620 bool CodeExtractor::isEligible() const { 621 if (Blocks.empty()) 622 return false; 623 BasicBlock *Header = *Blocks.begin(); 624 Function *F = Header->getParent(); 625 626 // For functions with varargs, check that varargs handling is only done in the 627 // outlined function, i.e vastart and vaend are only used in outlined blocks. 628 if (AllowVarArgs && F->getFunctionType()->isVarArg()) { 629 auto containsVarArgIntrinsic = [](const Instruction &I) { 630 if (const CallInst *CI = dyn_cast<CallInst>(&I)) 631 if (const Function *Callee = CI->getCalledFunction()) 632 return Callee->getIntrinsicID() == Intrinsic::vastart || 633 Callee->getIntrinsicID() == Intrinsic::vaend; 634 return false; 635 }; 636 637 for (auto &BB : *F) { 638 if (Blocks.count(&BB)) 639 continue; 640 if (llvm::any_of(BB, containsVarArgIntrinsic)) 641 return false; 642 } 643 } 644 return true; 645 } 646 647 void CodeExtractor::findInputsOutputs(ValueSet &Inputs, ValueSet &Outputs, 648 const ValueSet &SinkCands) const { 649 for (BasicBlock *BB : Blocks) { 650 // If a used value is defined outside the region, it's an input. If an 651 // instruction is used outside the region, it's an output. 652 for (Instruction &II : *BB) { 653 for (auto &OI : II.operands()) { 654 Value *V = OI; 655 if (!SinkCands.count(V) && definedInCaller(Blocks, V)) 656 Inputs.insert(V); 657 } 658 659 for (User *U : II.users()) 660 if (!definedInRegion(Blocks, U)) { 661 Outputs.insert(&II); 662 break; 663 } 664 } 665 } 666 } 667 668 /// severSplitPHINodesOfEntry - If a PHI node has multiple inputs from outside 669 /// of the region, we need to split the entry block of the region so that the 670 /// PHI node is easier to deal with. 671 void CodeExtractor::severSplitPHINodesOfEntry(BasicBlock *&Header) { 672 unsigned NumPredsFromRegion = 0; 673 unsigned NumPredsOutsideRegion = 0; 674 675 if (Header != &Header->getParent()->getEntryBlock()) { 676 PHINode *PN = dyn_cast<PHINode>(Header->begin()); 677 if (!PN) return; // No PHI nodes. 678 679 // If the header node contains any PHI nodes, check to see if there is more 680 // than one entry from outside the region. If so, we need to sever the 681 // header block into two. 682 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) 683 if (Blocks.count(PN->getIncomingBlock(i))) 684 ++NumPredsFromRegion; 685 else 686 ++NumPredsOutsideRegion; 687 688 // If there is one (or fewer) predecessor from outside the region, we don't 689 // need to do anything special. 690 if (NumPredsOutsideRegion <= 1) return; 691 } 692 693 // Otherwise, we need to split the header block into two pieces: one 694 // containing PHI nodes merging values from outside of the region, and a 695 // second that contains all of the code for the block and merges back any 696 // incoming values from inside of the region. 697 BasicBlock *NewBB = SplitBlock(Header, Header->getFirstNonPHI(), DT); 698 699 // We only want to code extract the second block now, and it becomes the new 700 // header of the region. 701 BasicBlock *OldPred = Header; 702 Blocks.remove(OldPred); 703 Blocks.insert(NewBB); 704 Header = NewBB; 705 706 // Okay, now we need to adjust the PHI nodes and any branches from within the 707 // region to go to the new header block instead of the old header block. 708 if (NumPredsFromRegion) { 709 PHINode *PN = cast<PHINode>(OldPred->begin()); 710 // Loop over all of the predecessors of OldPred that are in the region, 711 // changing them to branch to NewBB instead. 712 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) 713 if (Blocks.count(PN->getIncomingBlock(i))) { 714 Instruction *TI = PN->getIncomingBlock(i)->getTerminator(); 715 TI->replaceUsesOfWith(OldPred, NewBB); 716 } 717 718 // Okay, everything within the region is now branching to the right block, we 719 // just have to update the PHI nodes now, inserting PHI nodes into NewBB. 720 BasicBlock::iterator AfterPHIs; 721 for (AfterPHIs = OldPred->begin(); isa<PHINode>(AfterPHIs); ++AfterPHIs) { 722 PHINode *PN = cast<PHINode>(AfterPHIs); 723 // Create a new PHI node in the new region, which has an incoming value 724 // from OldPred of PN. 725 PHINode *NewPN = PHINode::Create(PN->getType(), 1 + NumPredsFromRegion, 726 PN->getName() + ".ce", &NewBB->front()); 727 PN->replaceAllUsesWith(NewPN); 728 NewPN->addIncoming(PN, OldPred); 729 730 // Loop over all of the incoming value in PN, moving them to NewPN if they 731 // are from the extracted region. 732 for (unsigned i = 0; i != PN->getNumIncomingValues(); ++i) { 733 if (Blocks.count(PN->getIncomingBlock(i))) { 734 NewPN->addIncoming(PN->getIncomingValue(i), PN->getIncomingBlock(i)); 735 PN->removeIncomingValue(i); 736 --i; 737 } 738 } 739 } 740 } 741 } 742 743 /// severSplitPHINodesOfExits - if PHI nodes in exit blocks have inputs from 744 /// outlined region, we split these PHIs on two: one with inputs from region 745 /// and other with remaining incoming blocks; then first PHIs are placed in 746 /// outlined region. 747 void CodeExtractor::severSplitPHINodesOfExits( 748 const SmallPtrSetImpl<BasicBlock *> &Exits) { 749 for (BasicBlock *ExitBB : Exits) { 750 BasicBlock *NewBB = nullptr; 751 752 for (PHINode &PN : ExitBB->phis()) { 753 // Find all incoming values from the outlining region. 754 SmallVector<unsigned, 2> IncomingVals; 755 for (unsigned i = 0; i < PN.getNumIncomingValues(); ++i) 756 if (Blocks.count(PN.getIncomingBlock(i))) 757 IncomingVals.push_back(i); 758 759 // Do not process PHI if there is one (or fewer) predecessor from region. 760 // If PHI has exactly one predecessor from region, only this one incoming 761 // will be replaced on codeRepl block, so it should be safe to skip PHI. 762 if (IncomingVals.size() <= 1) 763 continue; 764 765 // Create block for new PHIs and add it to the list of outlined if it 766 // wasn't done before. 767 if (!NewBB) { 768 NewBB = BasicBlock::Create(ExitBB->getContext(), 769 ExitBB->getName() + ".split", 770 ExitBB->getParent(), ExitBB); 771 SmallVector<BasicBlock *, 4> Preds(predecessors(ExitBB)); 772 for (BasicBlock *PredBB : Preds) 773 if (Blocks.count(PredBB)) 774 PredBB->getTerminator()->replaceUsesOfWith(ExitBB, NewBB); 775 BranchInst::Create(ExitBB, NewBB); 776 Blocks.insert(NewBB); 777 } 778 779 // Split this PHI. 780 PHINode *NewPN = 781 PHINode::Create(PN.getType(), IncomingVals.size(), 782 PN.getName() + ".ce", NewBB->getFirstNonPHI()); 783 for (unsigned i : IncomingVals) 784 NewPN->addIncoming(PN.getIncomingValue(i), PN.getIncomingBlock(i)); 785 for (unsigned i : reverse(IncomingVals)) 786 PN.removeIncomingValue(i, false); 787 PN.addIncoming(NewPN, NewBB); 788 } 789 } 790 } 791 792 void CodeExtractor::splitReturnBlocks() { 793 for (BasicBlock *Block : Blocks) 794 if (ReturnInst *RI = dyn_cast<ReturnInst>(Block->getTerminator())) { 795 BasicBlock *New = 796 Block->splitBasicBlock(RI->getIterator(), Block->getName() + ".ret"); 797 if (DT) { 798 // Old dominates New. New node dominates all other nodes dominated 799 // by Old. 800 DomTreeNode *OldNode = DT->getNode(Block); 801 SmallVector<DomTreeNode *, 8> Children(OldNode->begin(), 802 OldNode->end()); 803 804 DomTreeNode *NewNode = DT->addNewBlock(New, Block); 805 806 for (DomTreeNode *I : Children) 807 DT->changeImmediateDominator(I, NewNode); 808 } 809 } 810 } 811 812 /// constructFunction - make a function based on inputs and outputs, as follows: 813 /// f(in0, ..., inN, out0, ..., outN) 814 Function *CodeExtractor::constructFunction(const ValueSet &inputs, 815 const ValueSet &outputs, 816 BasicBlock *header, 817 BasicBlock *newRootNode, 818 BasicBlock *newHeader, 819 Function *oldFunction, 820 Module *M) { 821 LLVM_DEBUG(dbgs() << "inputs: " << inputs.size() << "\n"); 822 LLVM_DEBUG(dbgs() << "outputs: " << outputs.size() << "\n"); 823 824 // This function returns unsigned, outputs will go back by reference. 825 switch (NumExitBlocks) { 826 case 0: 827 case 1: RetTy = Type::getVoidTy(header->getContext()); break; 828 case 2: RetTy = Type::getInt1Ty(header->getContext()); break; 829 default: RetTy = Type::getInt16Ty(header->getContext()); break; 830 } 831 832 std::vector<Type *> paramTy; 833 834 // Add the types of the input values to the function's argument list 835 for (Value *value : inputs) { 836 LLVM_DEBUG(dbgs() << "value used in func: " << *value << "\n"); 837 paramTy.push_back(value->getType()); 838 } 839 840 // Add the types of the output values to the function's argument list. 841 for (Value *output : outputs) { 842 LLVM_DEBUG(dbgs() << "instr used in func: " << *output << "\n"); 843 if (AggregateArgs) 844 paramTy.push_back(output->getType()); 845 else 846 paramTy.push_back(PointerType::getUnqual(output->getType())); 847 } 848 849 LLVM_DEBUG({ 850 dbgs() << "Function type: " << *RetTy << " f("; 851 for (Type *i : paramTy) 852 dbgs() << *i << ", "; 853 dbgs() << ")\n"; 854 }); 855 856 StructType *StructTy = nullptr; 857 if (AggregateArgs && (inputs.size() + outputs.size() > 0)) { 858 StructTy = StructType::get(M->getContext(), paramTy); 859 paramTy.clear(); 860 paramTy.push_back(PointerType::getUnqual(StructTy)); 861 } 862 FunctionType *funcType = 863 FunctionType::get(RetTy, paramTy, 864 AllowVarArgs && oldFunction->isVarArg()); 865 866 std::string SuffixToUse = 867 Suffix.empty() 868 ? (header->getName().empty() ? "extracted" : header->getName().str()) 869 : Suffix; 870 // Create the new function 871 Function *newFunction = Function::Create( 872 funcType, GlobalValue::InternalLinkage, oldFunction->getAddressSpace(), 873 oldFunction->getName() + "." + SuffixToUse, M); 874 // If the old function is no-throw, so is the new one. 875 if (oldFunction->doesNotThrow()) 876 newFunction->setDoesNotThrow(); 877 878 // Inherit the uwtable attribute if we need to. 879 if (oldFunction->hasUWTable()) 880 newFunction->setHasUWTable(); 881 882 // Inherit all of the target dependent attributes and white-listed 883 // target independent attributes. 884 // (e.g. If the extracted region contains a call to an x86.sse 885 // instruction we need to make sure that the extracted region has the 886 // "target-features" attribute allowing it to be lowered. 887 // FIXME: This should be changed to check to see if a specific 888 // attribute can not be inherited. 889 for (const auto &Attr : oldFunction->getAttributes().getFnAttributes()) { 890 if (Attr.isStringAttribute()) { 891 if (Attr.getKindAsString() == "thunk") 892 continue; 893 } else 894 switch (Attr.getKindAsEnum()) { 895 // Those attributes cannot be propagated safely. Explicitly list them 896 // here so we get a warning if new attributes are added. This list also 897 // includes non-function attributes. 898 case Attribute::Alignment: 899 case Attribute::AllocSize: 900 case Attribute::ArgMemOnly: 901 case Attribute::Builtin: 902 case Attribute::ByVal: 903 case Attribute::Convergent: 904 case Attribute::Dereferenceable: 905 case Attribute::DereferenceableOrNull: 906 case Attribute::InAlloca: 907 case Attribute::InReg: 908 case Attribute::InaccessibleMemOnly: 909 case Attribute::InaccessibleMemOrArgMemOnly: 910 case Attribute::JumpTable: 911 case Attribute::Naked: 912 case Attribute::Nest: 913 case Attribute::NoAlias: 914 case Attribute::NoBuiltin: 915 case Attribute::NoCapture: 916 case Attribute::NoMerge: 917 case Attribute::NoReturn: 918 case Attribute::NoSync: 919 case Attribute::NoUndef: 920 case Attribute::None: 921 case Attribute::NonNull: 922 case Attribute::Preallocated: 923 case Attribute::ReadNone: 924 case Attribute::ReadOnly: 925 case Attribute::Returned: 926 case Attribute::ReturnsTwice: 927 case Attribute::SExt: 928 case Attribute::Speculatable: 929 case Attribute::StackAlignment: 930 case Attribute::StructRet: 931 case Attribute::SwiftError: 932 case Attribute::SwiftSelf: 933 case Attribute::WillReturn: 934 case Attribute::WriteOnly: 935 case Attribute::ZExt: 936 case Attribute::ImmArg: 937 case Attribute::ByRef: 938 case Attribute::EndAttrKinds: 939 case Attribute::EmptyKey: 940 case Attribute::TombstoneKey: 941 continue; 942 // Those attributes should be safe to propagate to the extracted function. 943 case Attribute::AlwaysInline: 944 case Attribute::Cold: 945 case Attribute::Hot: 946 case Attribute::NoRecurse: 947 case Attribute::InlineHint: 948 case Attribute::MinSize: 949 case Attribute::NoCallback: 950 case Attribute::NoDuplicate: 951 case Attribute::NoFree: 952 case Attribute::NoImplicitFloat: 953 case Attribute::NoInline: 954 case Attribute::NonLazyBind: 955 case Attribute::NoRedZone: 956 case Attribute::NoUnwind: 957 case Attribute::NullPointerIsValid: 958 case Attribute::OptForFuzzing: 959 case Attribute::OptimizeNone: 960 case Attribute::OptimizeForSize: 961 case Attribute::SafeStack: 962 case Attribute::ShadowCallStack: 963 case Attribute::SanitizeAddress: 964 case Attribute::SanitizeMemory: 965 case Attribute::SanitizeThread: 966 case Attribute::SanitizeHWAddress: 967 case Attribute::SanitizeMemTag: 968 case Attribute::SpeculativeLoadHardening: 969 case Attribute::StackProtect: 970 case Attribute::StackProtectReq: 971 case Attribute::StackProtectStrong: 972 case Attribute::StrictFP: 973 case Attribute::UWTable: 974 case Attribute::NoCfCheck: 975 case Attribute::MustProgress: 976 case Attribute::NoProfile: 977 break; 978 } 979 980 newFunction->addFnAttr(Attr); 981 } 982 newFunction->getBasicBlockList().push_back(newRootNode); 983 984 // Create an iterator to name all of the arguments we inserted. 985 Function::arg_iterator AI = newFunction->arg_begin(); 986 987 // Rewrite all users of the inputs in the extracted region to use the 988 // arguments (or appropriate addressing into struct) instead. 989 for (unsigned i = 0, e = inputs.size(); i != e; ++i) { 990 Value *RewriteVal; 991 if (AggregateArgs) { 992 Value *Idx[2]; 993 Idx[0] = Constant::getNullValue(Type::getInt32Ty(header->getContext())); 994 Idx[1] = ConstantInt::get(Type::getInt32Ty(header->getContext()), i); 995 Instruction *TI = newFunction->begin()->getTerminator(); 996 GetElementPtrInst *GEP = GetElementPtrInst::Create( 997 StructTy, &*AI, Idx, "gep_" + inputs[i]->getName(), TI); 998 RewriteVal = new LoadInst(StructTy->getElementType(i), GEP, 999 "loadgep_" + inputs[i]->getName(), TI); 1000 } else 1001 RewriteVal = &*AI++; 1002 1003 std::vector<User *> Users(inputs[i]->user_begin(), inputs[i]->user_end()); 1004 for (User *use : Users) 1005 if (Instruction *inst = dyn_cast<Instruction>(use)) 1006 if (Blocks.count(inst->getParent())) 1007 inst->replaceUsesOfWith(inputs[i], RewriteVal); 1008 } 1009 1010 // Set names for input and output arguments. 1011 if (!AggregateArgs) { 1012 AI = newFunction->arg_begin(); 1013 for (unsigned i = 0, e = inputs.size(); i != e; ++i, ++AI) 1014 AI->setName(inputs[i]->getName()); 1015 for (unsigned i = 0, e = outputs.size(); i != e; ++i, ++AI) 1016 AI->setName(outputs[i]->getName()+".out"); 1017 } 1018 1019 // Rewrite branches to basic blocks outside of the loop to new dummy blocks 1020 // within the new function. This must be done before we lose track of which 1021 // blocks were originally in the code region. 1022 std::vector<User *> Users(header->user_begin(), header->user_end()); 1023 for (auto &U : Users) 1024 // The BasicBlock which contains the branch is not in the region 1025 // modify the branch target to a new block 1026 if (Instruction *I = dyn_cast<Instruction>(U)) 1027 if (I->isTerminator() && I->getFunction() == oldFunction && 1028 !Blocks.count(I->getParent())) 1029 I->replaceUsesOfWith(header, newHeader); 1030 1031 return newFunction; 1032 } 1033 1034 /// Erase lifetime.start markers which reference inputs to the extraction 1035 /// region, and insert the referenced memory into \p LifetimesStart. 1036 /// 1037 /// The extraction region is defined by a set of blocks (\p Blocks), and a set 1038 /// of allocas which will be moved from the caller function into the extracted 1039 /// function (\p SunkAllocas). 1040 static void eraseLifetimeMarkersOnInputs(const SetVector<BasicBlock *> &Blocks, 1041 const SetVector<Value *> &SunkAllocas, 1042 SetVector<Value *> &LifetimesStart) { 1043 for (BasicBlock *BB : Blocks) { 1044 for (auto It = BB->begin(), End = BB->end(); It != End;) { 1045 auto *II = dyn_cast<IntrinsicInst>(&*It); 1046 ++It; 1047 if (!II || !II->isLifetimeStartOrEnd()) 1048 continue; 1049 1050 // Get the memory operand of the lifetime marker. If the underlying 1051 // object is a sunk alloca, or is otherwise defined in the extraction 1052 // region, the lifetime marker must not be erased. 1053 Value *Mem = II->getOperand(1)->stripInBoundsOffsets(); 1054 if (SunkAllocas.count(Mem) || definedInRegion(Blocks, Mem)) 1055 continue; 1056 1057 if (II->getIntrinsicID() == Intrinsic::lifetime_start) 1058 LifetimesStart.insert(Mem); 1059 II->eraseFromParent(); 1060 } 1061 } 1062 } 1063 1064 /// Insert lifetime start/end markers surrounding the call to the new function 1065 /// for objects defined in the caller. 1066 static void insertLifetimeMarkersSurroundingCall( 1067 Module *M, ArrayRef<Value *> LifetimesStart, ArrayRef<Value *> LifetimesEnd, 1068 CallInst *TheCall) { 1069 LLVMContext &Ctx = M->getContext(); 1070 auto Int8PtrTy = Type::getInt8PtrTy(Ctx); 1071 auto NegativeOne = ConstantInt::getSigned(Type::getInt64Ty(Ctx), -1); 1072 Instruction *Term = TheCall->getParent()->getTerminator(); 1073 1074 // The memory argument to a lifetime marker must be a i8*. Cache any bitcasts 1075 // needed to satisfy this requirement so they may be reused. 1076 DenseMap<Value *, Value *> Bitcasts; 1077 1078 // Emit lifetime markers for the pointers given in \p Objects. Insert the 1079 // markers before the call if \p InsertBefore, and after the call otherwise. 1080 auto insertMarkers = [&](Function *MarkerFunc, ArrayRef<Value *> Objects, 1081 bool InsertBefore) { 1082 for (Value *Mem : Objects) { 1083 assert((!isa<Instruction>(Mem) || cast<Instruction>(Mem)->getFunction() == 1084 TheCall->getFunction()) && 1085 "Input memory not defined in original function"); 1086 Value *&MemAsI8Ptr = Bitcasts[Mem]; 1087 if (!MemAsI8Ptr) { 1088 if (Mem->getType() == Int8PtrTy) 1089 MemAsI8Ptr = Mem; 1090 else 1091 MemAsI8Ptr = 1092 CastInst::CreatePointerCast(Mem, Int8PtrTy, "lt.cast", TheCall); 1093 } 1094 1095 auto Marker = CallInst::Create(MarkerFunc, {NegativeOne, MemAsI8Ptr}); 1096 if (InsertBefore) 1097 Marker->insertBefore(TheCall); 1098 else 1099 Marker->insertBefore(Term); 1100 } 1101 }; 1102 1103 if (!LifetimesStart.empty()) { 1104 auto StartFn = llvm::Intrinsic::getDeclaration( 1105 M, llvm::Intrinsic::lifetime_start, Int8PtrTy); 1106 insertMarkers(StartFn, LifetimesStart, /*InsertBefore=*/true); 1107 } 1108 1109 if (!LifetimesEnd.empty()) { 1110 auto EndFn = llvm::Intrinsic::getDeclaration( 1111 M, llvm::Intrinsic::lifetime_end, Int8PtrTy); 1112 insertMarkers(EndFn, LifetimesEnd, /*InsertBefore=*/false); 1113 } 1114 } 1115 1116 /// emitCallAndSwitchStatement - This method sets up the caller side by adding 1117 /// the call instruction, splitting any PHI nodes in the header block as 1118 /// necessary. 1119 CallInst *CodeExtractor::emitCallAndSwitchStatement(Function *newFunction, 1120 BasicBlock *codeReplacer, 1121 ValueSet &inputs, 1122 ValueSet &outputs) { 1123 // Emit a call to the new function, passing in: *pointer to struct (if 1124 // aggregating parameters), or plan inputs and allocated memory for outputs 1125 std::vector<Value *> params, StructValues, ReloadOutputs, Reloads; 1126 1127 Module *M = newFunction->getParent(); 1128 LLVMContext &Context = M->getContext(); 1129 const DataLayout &DL = M->getDataLayout(); 1130 CallInst *call = nullptr; 1131 1132 // Add inputs as params, or to be filled into the struct 1133 unsigned ArgNo = 0; 1134 SmallVector<unsigned, 1> SwiftErrorArgs; 1135 for (Value *input : inputs) { 1136 if (AggregateArgs) 1137 StructValues.push_back(input); 1138 else { 1139 params.push_back(input); 1140 if (input->isSwiftError()) 1141 SwiftErrorArgs.push_back(ArgNo); 1142 } 1143 ++ArgNo; 1144 } 1145 1146 // Create allocas for the outputs 1147 for (Value *output : outputs) { 1148 if (AggregateArgs) { 1149 StructValues.push_back(output); 1150 } else { 1151 AllocaInst *alloca = 1152 new AllocaInst(output->getType(), DL.getAllocaAddrSpace(), 1153 nullptr, output->getName() + ".loc", 1154 &codeReplacer->getParent()->front().front()); 1155 ReloadOutputs.push_back(alloca); 1156 params.push_back(alloca); 1157 } 1158 } 1159 1160 StructType *StructArgTy = nullptr; 1161 AllocaInst *Struct = nullptr; 1162 if (AggregateArgs && (inputs.size() + outputs.size() > 0)) { 1163 std::vector<Type *> ArgTypes; 1164 for (ValueSet::iterator v = StructValues.begin(), 1165 ve = StructValues.end(); v != ve; ++v) 1166 ArgTypes.push_back((*v)->getType()); 1167 1168 // Allocate a struct at the beginning of this function 1169 StructArgTy = StructType::get(newFunction->getContext(), ArgTypes); 1170 Struct = new AllocaInst(StructArgTy, DL.getAllocaAddrSpace(), nullptr, 1171 "structArg", 1172 &codeReplacer->getParent()->front().front()); 1173 params.push_back(Struct); 1174 1175 for (unsigned i = 0, e = inputs.size(); i != e; ++i) { 1176 Value *Idx[2]; 1177 Idx[0] = Constant::getNullValue(Type::getInt32Ty(Context)); 1178 Idx[1] = ConstantInt::get(Type::getInt32Ty(Context), i); 1179 GetElementPtrInst *GEP = GetElementPtrInst::Create( 1180 StructArgTy, Struct, Idx, "gep_" + StructValues[i]->getName()); 1181 codeReplacer->getInstList().push_back(GEP); 1182 new StoreInst(StructValues[i], GEP, codeReplacer); 1183 } 1184 } 1185 1186 // Emit the call to the function 1187 call = CallInst::Create(newFunction, params, 1188 NumExitBlocks > 1 ? "targetBlock" : ""); 1189 // Add debug location to the new call, if the original function has debug 1190 // info. In that case, the terminator of the entry block of the extracted 1191 // function contains the first debug location of the extracted function, 1192 // set in extractCodeRegion. 1193 if (codeReplacer->getParent()->getSubprogram()) { 1194 if (auto DL = newFunction->getEntryBlock().getTerminator()->getDebugLoc()) 1195 call->setDebugLoc(DL); 1196 } 1197 codeReplacer->getInstList().push_back(call); 1198 1199 // Set swifterror parameter attributes. 1200 for (unsigned SwiftErrArgNo : SwiftErrorArgs) { 1201 call->addParamAttr(SwiftErrArgNo, Attribute::SwiftError); 1202 newFunction->addParamAttr(SwiftErrArgNo, Attribute::SwiftError); 1203 } 1204 1205 Function::arg_iterator OutputArgBegin = newFunction->arg_begin(); 1206 unsigned FirstOut = inputs.size(); 1207 if (!AggregateArgs) 1208 std::advance(OutputArgBegin, inputs.size()); 1209 1210 // Reload the outputs passed in by reference. 1211 for (unsigned i = 0, e = outputs.size(); i != e; ++i) { 1212 Value *Output = nullptr; 1213 if (AggregateArgs) { 1214 Value *Idx[2]; 1215 Idx[0] = Constant::getNullValue(Type::getInt32Ty(Context)); 1216 Idx[1] = ConstantInt::get(Type::getInt32Ty(Context), FirstOut + i); 1217 GetElementPtrInst *GEP = GetElementPtrInst::Create( 1218 StructArgTy, Struct, Idx, "gep_reload_" + outputs[i]->getName()); 1219 codeReplacer->getInstList().push_back(GEP); 1220 Output = GEP; 1221 } else { 1222 Output = ReloadOutputs[i]; 1223 } 1224 LoadInst *load = new LoadInst(outputs[i]->getType(), Output, 1225 outputs[i]->getName() + ".reload", 1226 codeReplacer); 1227 Reloads.push_back(load); 1228 std::vector<User *> Users(outputs[i]->user_begin(), outputs[i]->user_end()); 1229 for (unsigned u = 0, e = Users.size(); u != e; ++u) { 1230 Instruction *inst = cast<Instruction>(Users[u]); 1231 if (!Blocks.count(inst->getParent())) 1232 inst->replaceUsesOfWith(outputs[i], load); 1233 } 1234 } 1235 1236 // Now we can emit a switch statement using the call as a value. 1237 SwitchInst *TheSwitch = 1238 SwitchInst::Create(Constant::getNullValue(Type::getInt16Ty(Context)), 1239 codeReplacer, 0, codeReplacer); 1240 1241 // Since there may be multiple exits from the original region, make the new 1242 // function return an unsigned, switch on that number. This loop iterates 1243 // over all of the blocks in the extracted region, updating any terminator 1244 // instructions in the to-be-extracted region that branch to blocks that are 1245 // not in the region to be extracted. 1246 std::map<BasicBlock *, BasicBlock *> ExitBlockMap; 1247 1248 unsigned switchVal = 0; 1249 for (BasicBlock *Block : Blocks) { 1250 Instruction *TI = Block->getTerminator(); 1251 for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i) 1252 if (!Blocks.count(TI->getSuccessor(i))) { 1253 BasicBlock *OldTarget = TI->getSuccessor(i); 1254 // add a new basic block which returns the appropriate value 1255 BasicBlock *&NewTarget = ExitBlockMap[OldTarget]; 1256 if (!NewTarget) { 1257 // If we don't already have an exit stub for this non-extracted 1258 // destination, create one now! 1259 NewTarget = BasicBlock::Create(Context, 1260 OldTarget->getName() + ".exitStub", 1261 newFunction); 1262 unsigned SuccNum = switchVal++; 1263 1264 Value *brVal = nullptr; 1265 switch (NumExitBlocks) { 1266 case 0: 1267 case 1: break; // No value needed. 1268 case 2: // Conditional branch, return a bool 1269 brVal = ConstantInt::get(Type::getInt1Ty(Context), !SuccNum); 1270 break; 1271 default: 1272 brVal = ConstantInt::get(Type::getInt16Ty(Context), SuccNum); 1273 break; 1274 } 1275 1276 ReturnInst::Create(Context, brVal, NewTarget); 1277 1278 // Update the switch instruction. 1279 TheSwitch->addCase(ConstantInt::get(Type::getInt16Ty(Context), 1280 SuccNum), 1281 OldTarget); 1282 } 1283 1284 // rewrite the original branch instruction with this new target 1285 TI->setSuccessor(i, NewTarget); 1286 } 1287 } 1288 1289 // Store the arguments right after the definition of output value. 1290 // This should be proceeded after creating exit stubs to be ensure that invoke 1291 // result restore will be placed in the outlined function. 1292 Function::arg_iterator OAI = OutputArgBegin; 1293 for (unsigned i = 0, e = outputs.size(); i != e; ++i) { 1294 auto *OutI = dyn_cast<Instruction>(outputs[i]); 1295 if (!OutI) 1296 continue; 1297 1298 // Find proper insertion point. 1299 BasicBlock::iterator InsertPt; 1300 // In case OutI is an invoke, we insert the store at the beginning in the 1301 // 'normal destination' BB. Otherwise we insert the store right after OutI. 1302 if (auto *InvokeI = dyn_cast<InvokeInst>(OutI)) 1303 InsertPt = InvokeI->getNormalDest()->getFirstInsertionPt(); 1304 else if (auto *Phi = dyn_cast<PHINode>(OutI)) 1305 InsertPt = Phi->getParent()->getFirstInsertionPt(); 1306 else 1307 InsertPt = std::next(OutI->getIterator()); 1308 1309 Instruction *InsertBefore = &*InsertPt; 1310 assert((InsertBefore->getFunction() == newFunction || 1311 Blocks.count(InsertBefore->getParent())) && 1312 "InsertPt should be in new function"); 1313 assert(OAI != newFunction->arg_end() && 1314 "Number of output arguments should match " 1315 "the amount of defined values"); 1316 if (AggregateArgs) { 1317 Value *Idx[2]; 1318 Idx[0] = Constant::getNullValue(Type::getInt32Ty(Context)); 1319 Idx[1] = ConstantInt::get(Type::getInt32Ty(Context), FirstOut + i); 1320 GetElementPtrInst *GEP = GetElementPtrInst::Create( 1321 StructArgTy, &*OAI, Idx, "gep_" + outputs[i]->getName(), 1322 InsertBefore); 1323 new StoreInst(outputs[i], GEP, InsertBefore); 1324 // Since there should be only one struct argument aggregating 1325 // all the output values, we shouldn't increment OAI, which always 1326 // points to the struct argument, in this case. 1327 } else { 1328 new StoreInst(outputs[i], &*OAI, InsertBefore); 1329 ++OAI; 1330 } 1331 } 1332 1333 // Now that we've done the deed, simplify the switch instruction. 1334 Type *OldFnRetTy = TheSwitch->getParent()->getParent()->getReturnType(); 1335 switch (NumExitBlocks) { 1336 case 0: 1337 // There are no successors (the block containing the switch itself), which 1338 // means that previously this was the last part of the function, and hence 1339 // this should be rewritten as a `ret' 1340 1341 // Check if the function should return a value 1342 if (OldFnRetTy->isVoidTy()) { 1343 ReturnInst::Create(Context, nullptr, TheSwitch); // Return void 1344 } else if (OldFnRetTy == TheSwitch->getCondition()->getType()) { 1345 // return what we have 1346 ReturnInst::Create(Context, TheSwitch->getCondition(), TheSwitch); 1347 } else { 1348 // Otherwise we must have code extracted an unwind or something, just 1349 // return whatever we want. 1350 ReturnInst::Create(Context, 1351 Constant::getNullValue(OldFnRetTy), TheSwitch); 1352 } 1353 1354 TheSwitch->eraseFromParent(); 1355 break; 1356 case 1: 1357 // Only a single destination, change the switch into an unconditional 1358 // branch. 1359 BranchInst::Create(TheSwitch->getSuccessor(1), TheSwitch); 1360 TheSwitch->eraseFromParent(); 1361 break; 1362 case 2: 1363 BranchInst::Create(TheSwitch->getSuccessor(1), TheSwitch->getSuccessor(2), 1364 call, TheSwitch); 1365 TheSwitch->eraseFromParent(); 1366 break; 1367 default: 1368 // Otherwise, make the default destination of the switch instruction be one 1369 // of the other successors. 1370 TheSwitch->setCondition(call); 1371 TheSwitch->setDefaultDest(TheSwitch->getSuccessor(NumExitBlocks)); 1372 // Remove redundant case 1373 TheSwitch->removeCase(SwitchInst::CaseIt(TheSwitch, NumExitBlocks-1)); 1374 break; 1375 } 1376 1377 // Insert lifetime markers around the reloads of any output values. The 1378 // allocas output values are stored in are only in-use in the codeRepl block. 1379 insertLifetimeMarkersSurroundingCall(M, ReloadOutputs, ReloadOutputs, call); 1380 1381 return call; 1382 } 1383 1384 void CodeExtractor::moveCodeToFunction(Function *newFunction) { 1385 Function *oldFunc = (*Blocks.begin())->getParent(); 1386 Function::BasicBlockListType &oldBlocks = oldFunc->getBasicBlockList(); 1387 Function::BasicBlockListType &newBlocks = newFunction->getBasicBlockList(); 1388 1389 for (BasicBlock *Block : Blocks) { 1390 // Delete the basic block from the old function, and the list of blocks 1391 oldBlocks.remove(Block); 1392 1393 // Insert this basic block into the new function 1394 newBlocks.push_back(Block); 1395 } 1396 } 1397 1398 void CodeExtractor::calculateNewCallTerminatorWeights( 1399 BasicBlock *CodeReplacer, 1400 DenseMap<BasicBlock *, BlockFrequency> &ExitWeights, 1401 BranchProbabilityInfo *BPI) { 1402 using Distribution = BlockFrequencyInfoImplBase::Distribution; 1403 using BlockNode = BlockFrequencyInfoImplBase::BlockNode; 1404 1405 // Update the branch weights for the exit block. 1406 Instruction *TI = CodeReplacer->getTerminator(); 1407 SmallVector<unsigned, 8> BranchWeights(TI->getNumSuccessors(), 0); 1408 1409 // Block Frequency distribution with dummy node. 1410 Distribution BranchDist; 1411 1412 SmallVector<BranchProbability, 4> EdgeProbabilities( 1413 TI->getNumSuccessors(), BranchProbability::getUnknown()); 1414 1415 // Add each of the frequencies of the successors. 1416 for (unsigned i = 0, e = TI->getNumSuccessors(); i < e; ++i) { 1417 BlockNode ExitNode(i); 1418 uint64_t ExitFreq = ExitWeights[TI->getSuccessor(i)].getFrequency(); 1419 if (ExitFreq != 0) 1420 BranchDist.addExit(ExitNode, ExitFreq); 1421 else 1422 EdgeProbabilities[i] = BranchProbability::getZero(); 1423 } 1424 1425 // Check for no total weight. 1426 if (BranchDist.Total == 0) { 1427 BPI->setEdgeProbability(CodeReplacer, EdgeProbabilities); 1428 return; 1429 } 1430 1431 // Normalize the distribution so that they can fit in unsigned. 1432 BranchDist.normalize(); 1433 1434 // Create normalized branch weights and set the metadata. 1435 for (unsigned I = 0, E = BranchDist.Weights.size(); I < E; ++I) { 1436 const auto &Weight = BranchDist.Weights[I]; 1437 1438 // Get the weight and update the current BFI. 1439 BranchWeights[Weight.TargetNode.Index] = Weight.Amount; 1440 BranchProbability BP(Weight.Amount, BranchDist.Total); 1441 EdgeProbabilities[Weight.TargetNode.Index] = BP; 1442 } 1443 BPI->setEdgeProbability(CodeReplacer, EdgeProbabilities); 1444 TI->setMetadata( 1445 LLVMContext::MD_prof, 1446 MDBuilder(TI->getContext()).createBranchWeights(BranchWeights)); 1447 } 1448 1449 /// Erase debug info intrinsics which refer to values in \p F but aren't in 1450 /// \p F. 1451 static void eraseDebugIntrinsicsWithNonLocalRefs(Function &F) { 1452 for (Instruction &I : instructions(F)) { 1453 SmallVector<DbgVariableIntrinsic *, 4> DbgUsers; 1454 findDbgUsers(DbgUsers, &I); 1455 for (DbgVariableIntrinsic *DVI : DbgUsers) 1456 if (DVI->getFunction() != &F) 1457 DVI->eraseFromParent(); 1458 } 1459 } 1460 1461 /// Fix up the debug info in the old and new functions by pointing line 1462 /// locations and debug intrinsics to the new subprogram scope, and by deleting 1463 /// intrinsics which point to values outside of the new function. 1464 static void fixupDebugInfoPostExtraction(Function &OldFunc, Function &NewFunc, 1465 CallInst &TheCall) { 1466 DISubprogram *OldSP = OldFunc.getSubprogram(); 1467 LLVMContext &Ctx = OldFunc.getContext(); 1468 1469 if (!OldSP) { 1470 // Erase any debug info the new function contains. 1471 stripDebugInfo(NewFunc); 1472 // Make sure the old function doesn't contain any non-local metadata refs. 1473 eraseDebugIntrinsicsWithNonLocalRefs(NewFunc); 1474 return; 1475 } 1476 1477 // Create a subprogram for the new function. Leave out a description of the 1478 // function arguments, as the parameters don't correspond to anything at the 1479 // source level. 1480 assert(OldSP->getUnit() && "Missing compile unit for subprogram"); 1481 DIBuilder DIB(*OldFunc.getParent(), /*AllowUnresolved=*/false, 1482 OldSP->getUnit()); 1483 auto SPType = DIB.createSubroutineType(DIB.getOrCreateTypeArray(None)); 1484 DISubprogram::DISPFlags SPFlags = DISubprogram::SPFlagDefinition | 1485 DISubprogram::SPFlagOptimized | 1486 DISubprogram::SPFlagLocalToUnit; 1487 auto NewSP = DIB.createFunction( 1488 OldSP->getUnit(), NewFunc.getName(), NewFunc.getName(), OldSP->getFile(), 1489 /*LineNo=*/0, SPType, /*ScopeLine=*/0, DINode::FlagZero, SPFlags); 1490 NewFunc.setSubprogram(NewSP); 1491 1492 // Debug intrinsics in the new function need to be updated in one of two 1493 // ways: 1494 // 1) They need to be deleted, because they describe a value in the old 1495 // function. 1496 // 2) They need to point to fresh metadata, e.g. because they currently 1497 // point to a variable in the wrong scope. 1498 SmallDenseMap<DINode *, DINode *> RemappedMetadata; 1499 SmallVector<Instruction *, 4> DebugIntrinsicsToDelete; 1500 for (Instruction &I : instructions(NewFunc)) { 1501 auto *DII = dyn_cast<DbgInfoIntrinsic>(&I); 1502 if (!DII) 1503 continue; 1504 1505 // Point the intrinsic to a fresh label within the new function. 1506 if (auto *DLI = dyn_cast<DbgLabelInst>(&I)) { 1507 DILabel *OldLabel = DLI->getLabel(); 1508 DINode *&NewLabel = RemappedMetadata[OldLabel]; 1509 if (!NewLabel) 1510 NewLabel = DILabel::get(Ctx, NewSP, OldLabel->getName(), 1511 OldLabel->getFile(), OldLabel->getLine()); 1512 DLI->setArgOperand(0, MetadataAsValue::get(Ctx, NewLabel)); 1513 continue; 1514 } 1515 1516 // If the location isn't a constant or an instruction, delete the 1517 // intrinsic. 1518 auto *DVI = cast<DbgVariableIntrinsic>(DII); 1519 Value *Location = DVI->getVariableLocation(); 1520 if (!Location || 1521 (!isa<Constant>(Location) && !isa<Instruction>(Location))) { 1522 DebugIntrinsicsToDelete.push_back(DVI); 1523 continue; 1524 } 1525 1526 // If the variable location is an instruction but isn't in the new 1527 // function, delete the intrinsic. 1528 Instruction *LocationInst = dyn_cast<Instruction>(Location); 1529 if (LocationInst && LocationInst->getFunction() != &NewFunc) { 1530 DebugIntrinsicsToDelete.push_back(DVI); 1531 continue; 1532 } 1533 1534 // Point the intrinsic to a fresh variable within the new function. 1535 DILocalVariable *OldVar = DVI->getVariable(); 1536 DINode *&NewVar = RemappedMetadata[OldVar]; 1537 if (!NewVar) 1538 NewVar = DIB.createAutoVariable( 1539 NewSP, OldVar->getName(), OldVar->getFile(), OldVar->getLine(), 1540 OldVar->getType(), /*AlwaysPreserve=*/false, DINode::FlagZero, 1541 OldVar->getAlignInBits()); 1542 DVI->setArgOperand(1, MetadataAsValue::get(Ctx, NewVar)); 1543 } 1544 for (auto *DII : DebugIntrinsicsToDelete) 1545 DII->eraseFromParent(); 1546 DIB.finalizeSubprogram(NewSP); 1547 1548 // Fix up the scope information attached to the line locations in the new 1549 // function. 1550 for (Instruction &I : instructions(NewFunc)) { 1551 if (const DebugLoc &DL = I.getDebugLoc()) 1552 I.setDebugLoc(DILocation::get(Ctx, DL.getLine(), DL.getCol(), NewSP)); 1553 1554 // Loop info metadata may contain line locations. Fix them up. 1555 auto updateLoopInfoLoc = [&Ctx, 1556 NewSP](const DILocation &Loc) -> DILocation * { 1557 return DILocation::get(Ctx, Loc.getLine(), Loc.getColumn(), NewSP, 1558 nullptr); 1559 }; 1560 updateLoopMetadataDebugLocations(I, updateLoopInfoLoc); 1561 } 1562 if (!TheCall.getDebugLoc()) 1563 TheCall.setDebugLoc(DILocation::get(Ctx, 0, 0, OldSP)); 1564 1565 eraseDebugIntrinsicsWithNonLocalRefs(NewFunc); 1566 } 1567 1568 Function * 1569 CodeExtractor::extractCodeRegion(const CodeExtractorAnalysisCache &CEAC) { 1570 if (!isEligible()) 1571 return nullptr; 1572 1573 // Assumption: this is a single-entry code region, and the header is the first 1574 // block in the region. 1575 BasicBlock *header = *Blocks.begin(); 1576 Function *oldFunction = header->getParent(); 1577 1578 // Calculate the entry frequency of the new function before we change the root 1579 // block. 1580 BlockFrequency EntryFreq; 1581 if (BFI) { 1582 assert(BPI && "Both BPI and BFI are required to preserve profile info"); 1583 for (BasicBlock *Pred : predecessors(header)) { 1584 if (Blocks.count(Pred)) 1585 continue; 1586 EntryFreq += 1587 BFI->getBlockFreq(Pred) * BPI->getEdgeProbability(Pred, header); 1588 } 1589 } 1590 1591 // Remove @llvm.assume calls that will be moved to the new function from the 1592 // old function's assumption cache. 1593 for (BasicBlock *Block : Blocks) { 1594 for (auto It = Block->begin(), End = Block->end(); It != End;) { 1595 Instruction *I = &*It; 1596 ++It; 1597 1598 if (match(I, m_Intrinsic<Intrinsic::assume>())) { 1599 if (AC) 1600 AC->unregisterAssumption(cast<CallInst>(I)); 1601 I->eraseFromParent(); 1602 } 1603 } 1604 } 1605 1606 // If we have any return instructions in the region, split those blocks so 1607 // that the return is not in the region. 1608 splitReturnBlocks(); 1609 1610 // Calculate the exit blocks for the extracted region and the total exit 1611 // weights for each of those blocks. 1612 DenseMap<BasicBlock *, BlockFrequency> ExitWeights; 1613 SmallPtrSet<BasicBlock *, 1> ExitBlocks; 1614 for (BasicBlock *Block : Blocks) { 1615 for (succ_iterator SI = succ_begin(Block), SE = succ_end(Block); SI != SE; 1616 ++SI) { 1617 if (!Blocks.count(*SI)) { 1618 // Update the branch weight for this successor. 1619 if (BFI) { 1620 BlockFrequency &BF = ExitWeights[*SI]; 1621 BF += BFI->getBlockFreq(Block) * BPI->getEdgeProbability(Block, *SI); 1622 } 1623 ExitBlocks.insert(*SI); 1624 } 1625 } 1626 } 1627 NumExitBlocks = ExitBlocks.size(); 1628 1629 // If we have to split PHI nodes of the entry or exit blocks, do so now. 1630 severSplitPHINodesOfEntry(header); 1631 severSplitPHINodesOfExits(ExitBlocks); 1632 1633 // This takes place of the original loop 1634 BasicBlock *codeReplacer = BasicBlock::Create(header->getContext(), 1635 "codeRepl", oldFunction, 1636 header); 1637 1638 // The new function needs a root node because other nodes can branch to the 1639 // head of the region, but the entry node of a function cannot have preds. 1640 BasicBlock *newFuncRoot = BasicBlock::Create(header->getContext(), 1641 "newFuncRoot"); 1642 auto *BranchI = BranchInst::Create(header); 1643 // If the original function has debug info, we have to add a debug location 1644 // to the new branch instruction from the artificial entry block. 1645 // We use the debug location of the first instruction in the extracted 1646 // blocks, as there is no other equivalent line in the source code. 1647 if (oldFunction->getSubprogram()) { 1648 any_of(Blocks, [&BranchI](const BasicBlock *BB) { 1649 return any_of(*BB, [&BranchI](const Instruction &I) { 1650 if (!I.getDebugLoc()) 1651 return false; 1652 BranchI->setDebugLoc(I.getDebugLoc()); 1653 return true; 1654 }); 1655 }); 1656 } 1657 newFuncRoot->getInstList().push_back(BranchI); 1658 1659 ValueSet inputs, outputs, SinkingCands, HoistingCands; 1660 BasicBlock *CommonExit = nullptr; 1661 findAllocas(CEAC, SinkingCands, HoistingCands, CommonExit); 1662 assert(HoistingCands.empty() || CommonExit); 1663 1664 // Find inputs to, outputs from the code region. 1665 findInputsOutputs(inputs, outputs, SinkingCands); 1666 1667 // Now sink all instructions which only have non-phi uses inside the region. 1668 // Group the allocas at the start of the block, so that any bitcast uses of 1669 // the allocas are well-defined. 1670 AllocaInst *FirstSunkAlloca = nullptr; 1671 for (auto *II : SinkingCands) { 1672 if (auto *AI = dyn_cast<AllocaInst>(II)) { 1673 AI->moveBefore(*newFuncRoot, newFuncRoot->getFirstInsertionPt()); 1674 if (!FirstSunkAlloca) 1675 FirstSunkAlloca = AI; 1676 } 1677 } 1678 assert((SinkingCands.empty() || FirstSunkAlloca) && 1679 "Did not expect a sink candidate without any allocas"); 1680 for (auto *II : SinkingCands) { 1681 if (!isa<AllocaInst>(II)) { 1682 cast<Instruction>(II)->moveAfter(FirstSunkAlloca); 1683 } 1684 } 1685 1686 if (!HoistingCands.empty()) { 1687 auto *HoistToBlock = findOrCreateBlockForHoisting(CommonExit); 1688 Instruction *TI = HoistToBlock->getTerminator(); 1689 for (auto *II : HoistingCands) 1690 cast<Instruction>(II)->moveBefore(TI); 1691 } 1692 1693 // Collect objects which are inputs to the extraction region and also 1694 // referenced by lifetime start markers within it. The effects of these 1695 // markers must be replicated in the calling function to prevent the stack 1696 // coloring pass from merging slots which store input objects. 1697 ValueSet LifetimesStart; 1698 eraseLifetimeMarkersOnInputs(Blocks, SinkingCands, LifetimesStart); 1699 1700 // Construct new function based on inputs/outputs & add allocas for all defs. 1701 Function *newFunction = 1702 constructFunction(inputs, outputs, header, newFuncRoot, codeReplacer, 1703 oldFunction, oldFunction->getParent()); 1704 1705 // Update the entry count of the function. 1706 if (BFI) { 1707 auto Count = BFI->getProfileCountFromFreq(EntryFreq.getFrequency()); 1708 if (Count.hasValue()) 1709 newFunction->setEntryCount( 1710 ProfileCount(Count.getValue(), Function::PCT_Real)); // FIXME 1711 BFI->setBlockFreq(codeReplacer, EntryFreq.getFrequency()); 1712 } 1713 1714 CallInst *TheCall = 1715 emitCallAndSwitchStatement(newFunction, codeReplacer, inputs, outputs); 1716 1717 moveCodeToFunction(newFunction); 1718 1719 // Replicate the effects of any lifetime start/end markers which referenced 1720 // input objects in the extraction region by placing markers around the call. 1721 insertLifetimeMarkersSurroundingCall( 1722 oldFunction->getParent(), LifetimesStart.getArrayRef(), {}, TheCall); 1723 1724 // Propagate personality info to the new function if there is one. 1725 if (oldFunction->hasPersonalityFn()) 1726 newFunction->setPersonalityFn(oldFunction->getPersonalityFn()); 1727 1728 // Update the branch weights for the exit block. 1729 if (BFI && NumExitBlocks > 1) 1730 calculateNewCallTerminatorWeights(codeReplacer, ExitWeights, BPI); 1731 1732 // Loop over all of the PHI nodes in the header and exit blocks, and change 1733 // any references to the old incoming edge to be the new incoming edge. 1734 for (BasicBlock::iterator I = header->begin(); isa<PHINode>(I); ++I) { 1735 PHINode *PN = cast<PHINode>(I); 1736 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) 1737 if (!Blocks.count(PN->getIncomingBlock(i))) 1738 PN->setIncomingBlock(i, newFuncRoot); 1739 } 1740 1741 for (BasicBlock *ExitBB : ExitBlocks) 1742 for (PHINode &PN : ExitBB->phis()) { 1743 Value *IncomingCodeReplacerVal = nullptr; 1744 for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) { 1745 // Ignore incoming values from outside of the extracted region. 1746 if (!Blocks.count(PN.getIncomingBlock(i))) 1747 continue; 1748 1749 // Ensure that there is only one incoming value from codeReplacer. 1750 if (!IncomingCodeReplacerVal) { 1751 PN.setIncomingBlock(i, codeReplacer); 1752 IncomingCodeReplacerVal = PN.getIncomingValue(i); 1753 } else 1754 assert(IncomingCodeReplacerVal == PN.getIncomingValue(i) && 1755 "PHI has two incompatbile incoming values from codeRepl"); 1756 } 1757 } 1758 1759 fixupDebugInfoPostExtraction(*oldFunction, *newFunction, *TheCall); 1760 1761 // Mark the new function `noreturn` if applicable. Terminators which resume 1762 // exception propagation are treated as returning instructions. This is to 1763 // avoid inserting traps after calls to outlined functions which unwind. 1764 bool doesNotReturn = none_of(*newFunction, [](const BasicBlock &BB) { 1765 const Instruction *Term = BB.getTerminator(); 1766 return isa<ReturnInst>(Term) || isa<ResumeInst>(Term); 1767 }); 1768 if (doesNotReturn) 1769 newFunction->setDoesNotReturn(); 1770 1771 LLVM_DEBUG(if (verifyFunction(*newFunction, &errs())) { 1772 newFunction->dump(); 1773 report_fatal_error("verification of newFunction failed!"); 1774 }); 1775 LLVM_DEBUG(if (verifyFunction(*oldFunction)) 1776 report_fatal_error("verification of oldFunction failed!")); 1777 LLVM_DEBUG(if (AC && verifyAssumptionCache(*oldFunction, *newFunction, AC)) 1778 report_fatal_error("Stale Asumption cache for old Function!")); 1779 return newFunction; 1780 } 1781 1782 bool CodeExtractor::verifyAssumptionCache(const Function &OldFunc, 1783 const Function &NewFunc, 1784 AssumptionCache *AC) { 1785 for (auto AssumeVH : AC->assumptions()) { 1786 auto *I = dyn_cast_or_null<CallInst>(AssumeVH); 1787 if (!I) 1788 continue; 1789 1790 // There shouldn't be any llvm.assume intrinsics in the new function. 1791 if (I->getFunction() != &OldFunc) 1792 return true; 1793 1794 // There shouldn't be any stale affected values in the assumption cache 1795 // that were previously in the old function, but that have now been moved 1796 // to the new function. 1797 for (auto AffectedValVH : AC->assumptionsFor(I->getOperand(0))) { 1798 auto *AffectedCI = dyn_cast_or_null<CallInst>(AffectedValVH); 1799 if (!AffectedCI) 1800 continue; 1801 if (AffectedCI->getFunction() != &OldFunc) 1802 return true; 1803 auto *AssumedInst = cast<Instruction>(AffectedCI->getOperand(0)); 1804 if (AssumedInst->getFunction() != &OldFunc) 1805 return true; 1806 } 1807 } 1808 return false; 1809 } 1810